Improving spatial distribution estimation of forest biomass with geostatistics: A case study for Rondonia, Brazil

Abstract

Mapping aboveground forest biomass is of fundamental importance for estimating CO2 emissions due to land use and land cover changes in the Brazilian Amazon. However, existing biomass maps for this region diverge in terms of the total biomass estimates derived, as well as in the spatial patterns of mapped biomass. In addition, no regional or location-specific measure of reliability accompanies most of these maps. In this study, 330 one-hectare plots from the RADAMBRASIL survey, acquired over and along areas adjacent to the state of Rondonia, were used to generate a biomass map over the entire region ˆ using geostatistics. The RADAMBRASIL samples were used to generate a biomass map, along with a measure of reliability for each biomass estimate at each location, using kriging with external drift with elevation, vegetation type and soil texture considered as biomass predictor variables. Cross-validation was performed using the sample plots to compare the performance of kriging against a simple biomass estimation using the sample mean. Overall, biomass varied from 225 to 486 Mg ha−1, with a local standard deviation ranging from 62 to 202 Mg ha−1. Large uncertainty values were obtained for regions with low sampling density, in particular in savanna areas. The geostatistical method adopted in this paper has the potential to be applied over the entire Brazilian Amazon region to provide more accurate local estimates of biomass, which would aid carbon flux estimation, along with measures of their reliability, and to identify areas where more sampling efforts should be concentrated.